Working memory allows individuals to access information that is no longer accessible from the sensorium. Decades of research indicate that distributed neural substrates support working memory processes (D’Esposito & Postle, 2015). Among the distributed cerebral cortices, the prefrontal cortex (PFC) and the posterior parietal cortex (PPC) have been most extensively studied. For example, the amplitude of evoked response in frontal and parietal cortices changes in response to varying working memory load (Braver et al., 1997, Carlson et al., 1998, Cohen et al., 1997, Jonides et al., 1997), and the capacity limit of working memory is reflected by saturated activity in the posterior parietal cortex (Todd and Marois, 2004, Xu and Chun, 2006). In addition to working memory load, cortical activity can be modulated by the types of stimuli maintained in working memory. For instance, when stimuli from different visual categories are presented, category-selective regions in the inferior temporal cortex, such as the fusiform face area (FFA) and the parahippocampal place area (PPA; O’Craven & Kanwisher, 2000), show increased activity scaled with working memory load for the respective categories (Druzgal and D’Esposito, 2001, Lepsien and Nobre, 2007, Ranganath et al., 2004). It is thought that these category-specific modulations are under the influence of top-down biasing signals from frontoparietal regions (D’Esposito & Postle, 2015), which enhance activity that encodes categorical working memory information to sustain working memory representation (Hwang et al., 2019, Hwang et al., 2020, Lee and D’Esposito, 2012, Miller et al., 2011, Miller and D’Esposito, 2005).

However, cortico-cortical projections alone are not sufficient to mediate the effects of top-down biasing, and subcortical regions likely have important contributions. Past studies using animal models have shown that the thalamus gates sensory processing (Halassa and Kastner, 2017, McAlonan et al., 2008) and regulates synchronization between occipito-temporal regions according to attentional demands (Saalmann et al., 2012). Lesioning the thalamus significantly reduces biasing effects that presumably originate from frontoparietal regions (Wimmer et al., 2015, Zhou et al., 2016). The thalamus consists of several nuclei, each has reciprocal connections with specific cortical regions. The anterior thalamus receives hippocampal inputs and reciprocally connects with the medial frontal cortex including the anterior cingulate cortex (Shibata, 1993, Shibata and Naito, 2005). The ventral lateral thalamus (VL) receives afferent inputs from the globus pallidus and the cerebellum, and in turn projects to the motor and premotor cortices (Schell and Strick, 1984, Strick, 1976). Pulvinar is the largest nucleus in the thalamus and has distributed connections with frontal, parietal, temporal cortices and the extrastriate visual area (Grieve et al., 2000, Jones et al., 1979, Robinson and Petersen, 1992, Yeterian and Pandya, 1985). Anatomically, MD has dense reciprocal connections with both medial and lateral PFC (Goldman-Rakic and Porrino, 1985, Jones, 2012). In rodents, elevated spiking activities in medial frontal neurons during the working memory delay period depend on inputs from the MD (Bolkan et al., 2017), and in humans, deep brain stimulation (DBS) of the MD increases working-memory-related errors (Peräkylä et al., 2017), providing causal evidence of thalamic involvement in working memory in humans. Together, these findings suggest that the human thalamus likely contributes to top-down biasing processes for working memory.

Two important sources of information can improve our understanding of how the human thalamus is related to working memory function. First, the human thalamus has a complex anatomy that consists of many distinct nuclei, each with a unique connectivity profile with different cortical regions. Selective nuclei, such as those found in the MD, have been extensively studied in the animal literature, yet how different subnuclei in the human thalamus are involved with working memory is not well understood. Second, thalamocortical interaction is likely critical for mediating the effects of top-down biasing for working memory, yet it is unclear whether task-evoked activity in the human thalamus interacts with cortical activity during working memory tasks.

The current study had two specific goals. First, to improve our understanding of the functional neuroanatomy of the human thalamus for working memory, we mapped the subdivision of the human thalamus in which activity is modulated by working memory load and categorical content of working memory information, in a well-powered sample. Second, to determine whether thalamocortical interactions support working memory processes, we adapted the previously-developed activity flow mapping approach (Cole et al., 2016, Cole et al., 2021) to test whether thalamocortical functional interaction can predict task-specific cortical activity. Accomplishing these two goals helped us identify specific sub-regions in the human thalamus and their associated thalamocortical interactions that are involved with working memory.